forked from jdkato/prose
-
Notifications
You must be signed in to change notification settings - Fork 0
/
summarize.go
227 lines (200 loc) · 6.54 KB
/
summarize.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
/*
Package summarize implements utilities for computing readability scores, usage statistics, and TL;DR summaries of text.
*/
package summarize
import (
"sort"
"strings"
"unicode"
"github.com/jdkato/prose/internal/util"
"github.com/jdkato/prose/tokenize"
"github.com/montanaflynn/stats"
)
// A Word represents a single word in a Document.
type Word struct {
Text string // the actual text
Syllables int // the number of syllables
}
// A Sentence represents a single sentence in a Document.
type Sentence struct {
Text string // the actual text
Length int // the number of words
Words []Word // the words in this sentence
Paragraph int
}
// A RankedParagraph is a paragraph ranked by its number of keywords.
type RankedParagraph struct {
Sentences []Sentence
Position int // the zero-based position within a Document
Rank int
}
// A Document represents a collection of text to be analyzed.
//
// A Document's calculations depend on its word and sentence tokenizers. You
// can use the defaults by invoking NewDocument, choose another implemention
// from the tokenize package, or use your own (as long as it implements the
// ProseTokenizer interface). For example,
//
// d := Document{Content: ..., WordTokenizer: ..., SentenceTokenizer: ...}
// d.Initialize()
type Document struct {
Content string // Actual text
NumCharacters float64 // Number of Characters
NumComplexWords float64 // PolysylWords without common suffixes
NumParagraphs float64 // Number of paragraphs
NumPolysylWords float64 // Number of words with > 2 syllables
NumSentences float64 // Number of sentences
NumSyllables float64 // Number of syllables
NumWords float64 // Number of words
Sentences []Sentence // the Document's sentences
WordFrequency map[string]int // [word]frequency
SentenceTokenizer tokenize.ProseTokenizer
WordTokenizer tokenize.ProseTokenizer
}
// An Assessment provides comprehensive access to a Document's metrics.
type Assessment struct {
// assessments returning an estimated grade level
AutomatedReadability float64
ColemanLiau float64
FleschKincaid float64
GunningFog float64
SMOG float64
// mean & standard deviation of the above estimated grade levels
MeanGradeLevel float64
StdDevGradeLevel float64
// assessments returning non-grade numerical scores
DaleChall float64
ReadingEase float64
}
// NewDocument is a Document constructor that takes a string as an argument. It
// then calculates the data necessary for computing readability and usage
// statistics.
//
// This is a convenience wrapper around the Document initialization process
// that defaults to using a WordBoundaryTokenizer and a PunktSentenceTokenizer
// as its word and sentence tokenizers, respectively.
func NewDocument(text string) *Document {
wTok := tokenize.NewWordBoundaryTokenizer()
sTok := tokenize.NewPunktSentenceTokenizer()
doc := Document{Content: text, WordTokenizer: wTok, SentenceTokenizer: sTok}
doc.Initialize()
return &doc
}
// Initialize calculates the data necessary for computing readability and usage
// statistics.
func (d *Document) Initialize() {
d.WordFrequency = make(map[string]int)
for i, paragraph := range strings.Split(d.Content, "\n\n") {
for _, s := range d.SentenceTokenizer.Tokenize(paragraph) {
wordCount := d.NumWords
d.NumSentences++
words := []Word{}
for _, word := range d.WordTokenizer.Tokenize(s) {
word = strings.TrimSpace(word)
if len(word) == 0 {
continue
}
d.NumCharacters += countChars(word)
if _, found := d.WordFrequency[word]; found {
d.WordFrequency[word]++
} else {
d.WordFrequency[word] = 1
}
syllables := Syllables(word)
words = append(words, Word{Text: word, Syllables: syllables})
d.NumSyllables += float64(syllables)
if syllables > 2 {
d.NumPolysylWords++
}
if isComplex(word, syllables) {
d.NumComplexWords++
}
d.NumWords++
}
d.Sentences = append(d.Sentences, Sentence{
Text: strings.TrimSpace(s),
Length: int(d.NumWords - wordCount),
Words: words,
Paragraph: i})
}
d.NumParagraphs++
}
}
// Assess returns an Assessment for the Document d.
func (d *Document) Assess() *Assessment {
a := Assessment{
FleschKincaid: d.FleschKincaid(), ReadingEase: d.FleschReadingEase(),
GunningFog: d.GunningFog(), SMOG: d.SMOG(), DaleChall: d.DaleChall(),
AutomatedReadability: d.AutomatedReadability(), ColemanLiau: d.ColemanLiau()}
gradeScores := []float64{
a.FleschKincaid, a.AutomatedReadability, a.GunningFog, a.SMOG,
a.ColemanLiau}
mean, merr := stats.Mean(gradeScores)
stdDev, serr := stats.StandardDeviation(gradeScores)
if merr != nil || serr != nil {
a.MeanGradeLevel = 0.0
a.StdDevGradeLevel = 0.0
} else {
a.MeanGradeLevel = mean
a.StdDevGradeLevel = stdDev
}
return &a
}
// Summary returns a Document's n highest ranked paragraphs according to
// keyword frequency.
func (d *Document) Summary(n int) []RankedParagraph {
rankings := []RankedParagraph{}
scores := d.Keywords()
for i := 0; i < int(d.NumParagraphs); i++ {
p := RankedParagraph{Position: i}
rank := 0
size := 0
for _, s := range d.Sentences {
if s.Paragraph == i {
size += s.Length
for _, w := range s.Words {
if score, found := scores[w.Text]; found {
rank += score
}
}
p.Sentences = append(p.Sentences, s)
}
}
// Favor longer paragraphs, as they tend to be more informational.
p.Rank = (rank * size)
rankings = append(rankings, p)
}
// Sort by raking:
sort.Sort(byRank(rankings))
// Take the top-n paragraphs:
size := len(rankings)
if size > n {
rankings = rankings[size-n:]
}
// Sort by chronological position:
sort.Sort(byIndex(rankings))
return rankings
}
type byRank []RankedParagraph
func (s byRank) Len() int { return len(s) }
func (s byRank) Swap(i, j int) { s[i], s[j] = s[j], s[i] }
func (s byRank) Less(i, j int) bool { return s[i].Rank < s[j].Rank }
type byIndex []RankedParagraph
func (s byIndex) Len() int { return len(s) }
func (s byIndex) Swap(i, j int) { s[i], s[j] = s[j], s[i] }
func (s byIndex) Less(i, j int) bool { return s[i].Position < s[j].Position }
func isComplex(word string, syllables int) bool {
if util.HasAnySuffix(word, []string{"es", "ed", "ing"}) {
syllables--
}
return syllables > 2
}
func countChars(word string) float64 {
count := 0
for _, c := range word {
if unicode.IsLetter(c) || unicode.IsNumber(c) {
count++
}
}
return float64(count)
}